Measure Impact
Lectures are provided as Jupyter Notebooks. The Impact Engine allows us to put causal inference methods into practice.
We follow Causal Inference: The Mixtape by Scott Cunningham as our foundational reference. All lectures consist of two parts. First, we cover the theory from the book. Second, we apply these concepts to product data and business decision-making contexts.
Causal Inference: The Mixtape
The material is organized in three parts. We begin with foundational causal models that clarify what causal effects mean and under which assumptions they are identified. We then study methods that rely on selection on observables, followed by methods designed to address selection on unobservables.
Foundations
Potential Outcomes Model
We introduce the potential outcomes framework as the foundational model for causal inference. This section formalizes the fundamental problem of causal inference—missing counterfactuals—and explains why randomization resolves it. The goal is to establish a precise language for defining causal effects and understanding what can and cannot be identified from data.
Causal Graphical Models
We introduce directed acyclic graphs (DAGs) as a complementary representation of causal assumptions. DAGs provide a visual and formal tool for reasoning about identification, making explicit the roles of confounders, mediators, and colliders. This section emphasizes how graphical structure encodes assumptions rather than estimates.
Selection on Observables
Matching & Propensity Scores
We discuss methods for causal inference when selection into treatment depends only on observed covariates. This section covers matching estimators and propensity score methods, highlighting the conditions under which covariate balance is sufficient for identification.
Machine Learning for Causal Inference
We explore how machine learning methods can improve causal inference, focusing on high-dimensional covariate settings. Topics include LASSO-based covariate selection and double machine learning, with an emphasis on separating prediction from causal identification.
Pooled Ordinary Least Squares (OLS)
We discuss pooled OLS estimation with panel data and clarify the assumptions under which it can be interpreted causally. The section emphasizes when pooled regression is valid and when it fails due to unobserved heterogeneity or dynamic selection.
Selection on Unobservables
Instrumental Variables
We introduce instrumental variables as a method for causal inference when unobserved confounding is present. The focus is on identification assumptions, interpretation of local average treatment effects, and the role of instruments in causal inference.
Regression Discontinuity
We discuss regression discontinuity designs that exploit sharp or fuzzy thresholds in treatment assignment. The section emphasizes identification at the cutoff and the interpretation of local causal effects.
Difference-in-Differences
We present difference-in-differences methods for panel data, including the parallel trends assumption and recent methodological advances. Connections to selection models and potential outcomes are made explicit.
Synthetic Control
We discuss synthetic control methods for comparative case studies with panel data, focusing on transparent construction of counterfactuals and assumptions about factor structures and trends.